1 Introduction

The East Asian summer monsoon (EASM) is a distinctive component of the Asian climate system and plays a crucial role in the livelihood of more than 1.5 billion people (e.g. Zhou et al. 2009; Huang et al. 2012; Chen et al. 2019). The EASM exhibits intense interannual variability, and the induced severe droughts and floods can cause great economic loss and societal impacts (e.g. Huang and Sun 1992; Wang et al. 2001; Ding and Chan 2005; Sun et al. 2010). Hence, understanding the mechanisms of the EASM variation may greatly benefit those people inhabiting the region.

As an important component of the EASM system (Wang et al. 2008), the western North Pacific anticyclone (WNPAC) is closely coupled with the EASM rainfall (e.g., Huang and Wu 1989; Lu 2004). This anticyclone suppresses the convection over the western North Pacific (WNP) and transports water vapor to the mei-yu/baiu/changma region, thus forming a dipole rainfall pattern. El Niño-Southern Oscillation (ENSO) is considered as a key factor for the interannual variability of EASM (e.g., Zhang et al. 1996; Wu et al. 2003; Feng et al. 2011; Chen et al. 2013b). It has been revealed that a strengthened WNPAC tends to appear in the decaying phase of El Niño (Wang et al. 2008), and this anomalous WNPAC is closely connected to the North Indian Ocean (NIO) sea surface temperature (SST) anomalies (Xie et al. 2009; Wu et al.2009) and tropical North Atlantic anomalies (TNA) SST (Rong et al. 2010; Hong et al. 2014; Li et al. 2016; Wang et al.2017). Xie et al. (2009) demonstrated that the Indian Ocean SST anomalies play an important role in prolonging the impact of winter El Niño on the following summer WNPAC via the so-called “Indian Ocean capacitor mechanism”. In addition, the SST warming in the TNA induced by ENSO also plays a crucial role after the late-1970s (Hong et al. 2014; Chang et al. 2016), through inducing east-west overturning vertical circulation and yielding Kelvin waves (Rong et al. 2010; Li et al. 2016; Wang et al. 2017).

The Coupled Model Inter-comparison Project (CMIP) provides a large amount of model output, the validation of which reflects the ability of current climate models and is an essential tool to study the monsoon variation. Through many years of development, the basic features and interannual variation of EASM can be captured well in the current models. For example, climate models in Phase 5 of the CMIP (CMIP5) showed improvements in the simulation of EASM climatology and interannual variability to the Phase 3 of the CMIP (CMIP3) models (Sperber et al. 2013; Feng et al. 2014; Song and Zhou 2014a; Gong et al. 2018). However, there is still much uncertainty in the EASM simulation for CMIP5 (Huang et al. 2013). Some common biases include a weak WNPAC and the underestimation of rainfall in the mei-yu/baiu/changma rainband (Song and Zhou 2014a; Gong et al. 2018). The EASM in the coupled models of CMIP5 and their corresponding atmospheric models have also been evaluated (Sperber et al. 2013; Song and Zhou 2014b). Song and Zhou (2014b) emphasized that air-sea coupling could improve the simulation of the interannual EASM pattern. Recently, the outputs from the latest climate models in Phase 6 of the CMIP (CMIP6) have been released (Eyring et al. 2016). Through comparing the ability of eight CMIP5 and CMIP6 models in simulating the summer rainfall in China and the EASM, Xin et al. (2020) stressed the importance of the SST anomalies over eastern Indian and the western Pacific Oceans in simulating EASM.

So far, the assessment of the climatology and interannual variability of the EASM in CMIP6 was not carried out systematically. It remains unknown as to whether the CMIP6 models can improve the simulation of the two key systems of EASM (i.e., mei-yu/baiu/changma rainband and WNPAC) compared to CMIP5. Additionally, the role of air-sea coupling in reducing the uncertainty of EASM simulation deserves further study. Therefore, based on comparison with CMIP5, the ability of current CMIP6 in modeling the climatological and interannual variation of EASM will be systematically examined in this study. We put more effort into exploring the simulation biases of CMIP5 and CMIP6 models in the EASM. Making clear these questions is beneficial to improve the simulation skill of models and precisely predict the interannual variability of EASM.

The remainder of this study is organized as follows: Sect. 2 describes the model, observational datasets, and methods. The climatology of EASM in CMIP5 and CMIP6 are examined and compared in Sect. 3. Section 4 displays the differences between CMIP5 and CMIP6 in the interannual variability of EASM, and the skill origins are further addressed. Finally, the summary is given in Sect. 5.

2 Data and methods

The first simulation (r1i1p1 for CMIP5 and r1i1p1f1 for CMIP6) of the historical experiments from 18 CMIP5 (Taylor et al. 2012) and 18 CMIP6 (Eyring et al. 2016) models are analyzed in this study, excluding CNRM-CM6-1 (r1i1p1f2) and UKESM1-0-LL (r1i1p1f2) in CMIP6, because of the lack of the first realization in the two models. The basic information of these models in CMIP5 and CMIP6 is listed in Tables 1 and 2, respectively. All the model data are bilinearly interpolated onto a 2.5° × 2.5° grid before analysis. The multi-model ensemble (MME) is calculated by averaging the variables in all the CMIP5 and CMIP6 models with equal weighting. The climatology and interannual variability of EASM are calculated for each model first, and then their MMEs are obtained by averaging with equal weight.

Table 1 18 CMIP5 models used in this study
Full size table
Table 2 18 CMIP6 models used in this study
Full size table

The observational proxies of the atmospheric variables are from the monthly mean data of ERA5 (Hersbach et al., 2020), developed by the European Center for Medium-range Forecasting (ECMWF). The dataset has a 0.25° × 0.25° horizontal resolution and spans the period from 1979 to present. The Climate Prediction Center Merged Analysis of Precipitation (CMAP; Xie and Arkin 20081997), which provide monthly precipitation anomalies in a 2.5° × 2.5° grid since 1979, are used as the global precipitation proxy. The monthly SST data used in this study are from the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST) dataset (Rayner et al. 2003), which has a horizontal resolution of 1° × 1° and covers the period from 1870 to the present.

A widely used EASM index (EASMI) proposed by Wang and Fan (1999) is selected to depict the intensity of the EASM, which is calculated as the difference in summer (June–July–August–mean, JJA) 850-hPa zonal winds between regions of 5°N–15°N, 90°E–130°E and 22.5°N–32.5°N, 110°E–140°E, but with the sign reversed. In this case, a positive EASMI stands for a strong EASM condition, which features more rainfall along the mei-yu/baiu/changma band. According to Wang et al. (2008), this index is highly correlated to the first principal component of EASM and can be considered as a simple representation of the dominant interannual variability mode of EASM. The interannual variability of EASM is obtained by regressing the precipitation and 850-hPa wind upon the EASMI in the observation and each model. The SST anomalies averaged over the region (5°S–5°N, 170°W–120°W) in December–January–February–mean (DJF) is used as the ENSO index, which is denoted as the Niño-3.4 index. This study focuses on the period from 1979 to 2005. All the data are detrended prior to the analysis. Seasonal means used throughout this study are constructed by averaging data for DJF, March–April–May (MAM), and JJA. The significance of MME is tested based on the consensus of the individual models. The MME is considered significant if more than 75% of the individual models agree on the sign of the anomalies (Power et al. 2012). The statistical significance of the differences in MME patterns between CMIP5 and CMIP6 models are estimated based on mean test using a two-tailed Student’s t-test.

To quantify the performance of the CMIP6 models and CMIP5 models, the skill score is calculated (Taylor 2001; Chen et al. 2013a), which combines the spatial distribution and magnitude together:

$${\text{Skill Score = }}\frac{{4{{(1{\text{+}}{R_{{\text{fr}}}})}^4}}}{{{{({S_{{\text{fr}}}}{\text{+}}1/{S_{{\text{fr}}}})}^2}{{(1+{R_0})}^4}}}$$
(1)

,

where Rfr is the field correlation between the observed pattern and each model pattern, R0 is the attainable maximum correlation (in this case, R0 = 1), and Sfr is the ratio of the model’s pattern standard deviation against the observation.

3 The climatology of EASM simulated by CMIP5 and CMIP6 models

In this section, we evaluated the models’ performance in simulating the summer atmospheric circulation and rainfall over East Asia (0°–50°N, 100°E–150°E) during the time period of 1979–2005. Figure 1 compares the climatology of precipitation and 850-hPa winds simulated by the CMIP5 and CMIP6 MMEs with the observation. In the observation, an anticyclonic circulation dominates the western Pacific together with the westerly winds over the Indo–china peninsula and South China Sea, bringing abundant water vapor into East Asia (Fig. 1a). Correspondingly, there are four rainfall centers in the tropics from western Indo–China peninsula to western Pacific, and one monsoon rainband extending from eastern China to Japan (i.e., mei-yu/baiu/changma). Both the CMIP5 and the CMIP6 MMEs capture these main features of the EASM (Fig. 1b, c). However, similar biases in both CMIP5 and CMIP6 MMEs are characterized by a cyclone over western Pacific, indicating a weakened western Pacific subtropical high (WPSH; Fig. 1d, e). In addition, the monsoon rainband and the rainfall centers along the tropics are relatively weaker than the observation, while the rainfall on the southwestern flank of the Tibetan Plateau is overestimated (Fig. 1d, e). The latter might be due to the treatment of the topography of the Tibetan Plateau. Such model biases in reproducing the magnitude of WPSH and the associated monsoon rainfall band are noted by previous studies (Zhou and Li 2002; Song and Zhou 2014a). It is worth mentioning that the westerly winds over the Indo–China peninsula is stronger in CMIP6 MME than that in CMIP5 MME, contributing to the stronger rainfall centers in the tropics (Fig. 1f). Moreover, the biases of the WPSH and mei-yu/baiu/changma rainband are reduced in the CMIP6 MME with a stronger EASM (Fig. 1f). The above differences between the CMIP5 and CMIP6 MMEs suggest improvements of CMIP6 models in reproducing the summertime climatological low-level winds and precipitation.

Fig. 1
figure 1

The climatological distribution of JJA precipitation (shaded; mm day− 1) and 850-hPa wind (vectors; m s− 1) in the (a) observation (OBS), (b) CMIP5 MME and (c) CMIP6 MME. The difference of precipitation and 850-hPa wind between (d) the CMIP5 MME and the observation, (e) the CMIP6 MME and the observation, (f) the CMIP6 MME and CMIP5 MME. Red winds and black dots in (f) indicate differences exceeding the 90% confidence level, as calculated by the Student’s t test

Full size image

The inter-model spreads estimated with the standard deviation among 18 CMIP5 and 18 CMIP6 models are shown in Fig. 2. For CMIP5 models, the spreads of 850-hPa zonal wind (U850) are large over the WNP, straddling the corresponding wind speed maximum over both the mei-yu/baiu/changma region and western Pacific (Fig. 2a). The result reflects that the simulated locations of the wind maximum and their intensities vary largely among models. The inter-model spreads of 850-hPa meridional wind (V850) are large over East Asia and north western Indo–China peninsula (Fig. 2b). Corresponding to the spreads of 850-hPa winds, the precipitation (Pr) varies largely in tropics from Indo–China peninsula to western Pacific (Fig. 2c). Similar features are also evident in CMIP6 models (Fig. 2d–f), but the magnitudes of these spreads are smaller in CMIP6 compared to those in CMIP5 models, suggesting the significant improvement of CMIP6 models in this aspect (Fig. 2 g–i).

Fig. 2
figure 2

The inter-model spread of JJA (a) 850-hPa zonal wind (U850; m s− 1), (b) 850-hPa meridional wind (V850; m s− 1), (c) precipitation (Pr; mm day− 1) for climatology among the CMIP5 models. (d)–(f), (g)–(i) same as in Figures (a)–(c) but for CMIP6 models, difference between the CMIP6 and CMIP5 models, respectively

Full size image

To delineate the performances of the CMIP5 and CMIP6 models in representing the EASM climatology in more detail, Taylor diagrams are plotted for the area of East Asia (0°–50°N, 100°E–150°E; Fig. 3). Taylor diagram is constructed based on the spatial correlation between the simulated and observed patterns, the normalized standard deviation of spatial pattern, and the centered pattern root-mean-square difference (Taylor et al. 2012). In general, the pattern correlations of U850, V850 and Pr between most of the models and observations range from 0.5 to 0.9, suggesting that CMIP5 and CMIP6 models reasonably simulate the spatial feature of the circulation and precipitation over the East Asia region, as seen also in Fig. 1. The standard deviations of the simulated Pr both in CMIP5 and CMIP6 are almost identical to those in the observations with relatively small inter-model spread, whereas the U850 and V850 in most models have greater standard deviations compared to the observations (Fig. 3b). In comparison, the MME of CMIP6 (Fig. 3b) outperforms the CMIP5 (Fig. 3a), and the inter-model spreads for CMIP6 (Fig. 3b) are much smaller than CMIP5 (Fig. 3a). Hence, we may conclude a better simulation of spatial feature of the circulation and precipitation over East Asia in the CMIP6 models than the CMIP5 models.

Fig. 3
figure 3

Taylor diagrams of JJA Pr (green), V850 (blue) and U850 (red) for climatology of (a) CMIP5 and (b) CMIP6 models in East Asia (0°–50°N, 100°E–150°E), respectively

Full size image

The Taylor diagram can reflect the resemblance of the spatial pattern but cannot describe the intensity of these variables (e.g., Gong et al. 2014). Thus, we further examined the skill scores of CMIP5 and CMIP6 models by using Eq. (1). Generally, the skill score of U850 is higher than those of V850 and Pr in most models, indicative of a better reproduction of zonal wind than meridional wind and precipitation (Fig. 4). Among 18 CMIP5 models (Fig. 4a), CNRM-CM5 performs best with the highest skill scores for the U850 (0.89) and V850 (0.72), which is consistent with the previous study of (2020). CNRM-CM5 also show reasonable precipitation performances with the skill score of 0.58. Among the CMIP6 models (Fig. 4b), CESM2 and FGOALS-f3-L show reasonable performances, as evidenced by high skill scores for the U850 (0.86; 0.84), V850 (0.75; 0.74) and Pr (0.59; 0.75) variables. In particular, the skills are generally improved from the CMIP5 to CMIP6 in the same modelling group. For example, the skill scores of U850 (V850; Pr) improve from 0.61 (0.57; 0.29) in MIROC5 to 0.75 (0.72; 0.56) in MIROC6. While there are also individual CMIP6 models that have not improved their abilities in some aspects compared with the previous CMIP5 models. For instance, both the skills of U850 and Pr are degraded from CNRM-CM5 (0.89; 0.58) to CNRM-CM6-1 (0.69; 0.47). In comparison, the skill scores of U850, V850, Pr in the MME are improved from CMIP5 (0.62; 0.44; 0.49) to CMIP6 (0.7; 0.56; 0.59). Moreover, the skill score ranges from 0.06 to 0.89 in CMIP5 models, while the skill score ranges from 0.28 to 0.86 in CMIP6 models. These results suggest that the climatological wind and precipitation over East Asia are significantly improved from CMIP5 to CMIP6 in the intensities.

Fig. 4
figure 4

Scatterplots of the skill score of JJA U850 (red), V850 (blue) and Pr (green) for climatology in (a) CMIP5 and (b) CMIP6 models. The skill is relative to ERA5 and CMAP over East Asia (0°–50°N, 100°E–150°E). The dashed lines indicate the skill score of MME

Full size image

4 The interannual variability of EASM simulated by CMIP5 and CMIP6 models

The precipitation and 850-hPa wind anomalies regressed on the EASMI are presented to evaluate the simulation of the interannual variability of EASM. To discuss the possible relationship between the simulation skill of climatology and interannual variability of EASM, Fig. 5 presents the scatterplots of the skill score of climatological precipitation versus the skill score of precipitation anomalies regressed upon the EASMI in CMIP5 and CMIP6 by using Eq. (1). It is clearly seen that the CMIP6 model with a higher skill in simulating the climatological precipitation tend to reproduce a more realistic anomalous precipitation related to the EASM, with the inter-model correlation coefficient of 0.53 which exceeding the 95% confidence level (Fig. 5b). However, the inter-model relationship among CMIP5 models is much weaker than that in CMIP6, with the correlation coefficient of 0.35 (Fig. 5a). These suggest that, compared with CMIP5, the model’s reproducibility of the observed interannual anomalous precipitation associated with EASM in CMIP6, is closely tied to the climatological precipitation in East Asia.

Fig. 5
figure 5

Scatterplots of the skill score of JJA precipitation for climatology vs. the skill score of precipitation anomalies regressed upon the EASMI in the (a) 18 CMIP5 and (b) 18 CMIP6 models. The skill is relative to ERA5 and CMAP over East Asia (0°–50°N, 100°E–150°E). Black lines and “r” denote linear fit

Full size image

The results of CMIP5 and CMIP6 MMEs are compared with the observation in Fig. 6. The observed wind anomalies show a strong anticyclone covering the entire WNP region, with increased southwesterly winds over southeastern China and westerly winds along the mei-yu/baiu/changma region (Fig. 6a). There also exists an anomalous cyclone to the north of the anomalous WNPAC during strong EASM epoch. Correspondingly, the precipitation anomalies exhibit the dipole pattern with a negative southern section over the WNP and a positive northern section including two centers along the mei-yu/baiu/changma region (Fig. 6a). Compared to the observation, both the CMIP5 and CMIP6 MMEs capture the main features of anomalous WNPAC and dipole rainfall pattern (Fig. 6b, c). But the simulated anomalous WNPAC is weaker, inducing the lower rainfall anomaly magnitude over the WNP and mei-yu/baiu/changma front (Fig. 6b, c). Moreover, the cyclonic wind anomalies over Japan are missing in the MMEs, which have also been identified in previous studies (Kosaka et al. 2012; Gong et al. 2018). Since the anomalous cyclone is located in the middle and high latitudes where the atmospheric processes are more chaotic than those in the low latitudes, it may be difficult to be well reproduced by the current climate models. Compared to CMIP5 MME, the anomalous circulation and precipitation biases in CMIP6 MME are alleviated (Fig. 6d). The anomalous WNPAC become strengthened in CMIP6 MME compared to that in CMIP5 MME. And the anomalous rainfall magnitudes over the mei-yu/baiu/changma and WNP regions are also enhanced (Fig. 6d).

Fig. 6
figure 6

The horizontal distribution of JJA precipitation (shaded; mm day− 1) and 850-hPa wind (vectors; m s− 1) regressed upon the EASMI in (a) observation, (b) CMIP5 MME, (c) CMIP6 MME and (d) the difference between CMIP6 MME and CMIP5 MME. Black dots indicate rainfall anomalies exceeding the 90% confidence level. Winds anomalies exceeding the 90% confidence level are shown (red) in a, b, c (d). The blue and red boxes in (a) indicate the mei-yu/baiu/changma (28°N–38°N, 105°E–150°E) and WNP (10°N–20°N, 105°E–150°E) regions, respectively

Full size image

Given the dominance of anomalous WNPAC on the water vapor transport and thereby monsoon rainfall associated with the EASM, Fig. 7 shows the model performance in simulating the intensity of anomalous WNPAC. An anomalous WNPAC index (WNPACI) is defined using the area-averaged vorticity over the region (10°N–30°N, 110°E–150°E), which is determined by the domain of anticyclone as shown in Fig. 6a. The negative WNPACI indicates an anomalous anticyclone, and a strengthened EASM. Most of the CMIP5 and CMIP6 models reproduce a muted WNPACI, indicating a relatively weak EASM (Fig. 7). There are 17 of 18 CMIP5 models and 11 of 18 CMIP6 models in which the WNPAC is weaker than the observation. The intensity of WNPACI is only − 0.98 s− 1 in the MME of CMIP5, but reaches to -1.22 s− 1 in the MME of CMIP6 which is more closer to the observation (-1.36 s− 1). Thus, there is an evident improvement in simulating the WNPACI from CMIP5 to CMIP6 both for the intensity and the number of models.

Fig. 7
figure 7

The area averaged precipitation (mm day− 1) and the WNPAC index (s− 1) regressed upon the EASMI in (a) CMIP5 and (b) CMIP6 models. The blue and red dots represent the mei-yu/baiu/changma and WNP regions (boxes shown in Fig. 6a), respectively. Grey bar represents the WNPACI defined by the vorticity averaged over (10°N–30°N, 110°E–150°E). The dashed lines indicate the observational magnitudes

Full size image

The area-averaged precipitation anomalies over the mei-yu/baiu/changma (28°N–38°N, 105°E–150°E) and WNP (10°N–20°N, 105°E–150°E) regions are also displayed in Fig. 7. In the observation, the magnitude of precipitation anomalies averaged over the WNP surpasses more than two times of that over the mei-yu/baiu/changma region. Most models reproduce the same signs of precipitation over the two regions as the observations, but underestimate the precipitation magnitude especially over the mei-yu/baiu/changma region. The weakened WNPACI underestimates the suppressed convection over the WNP and reduces the water vapor transport to the East Asia, therefore weakening the mei-yu/baiu/changma rainfall belt. However, consistent with the relatively stronger WNPACI in CMIP6 than CMIP5, the precipitation anomalies over the WNP and mei-yu/baiu/changma regions are also generally improved in CMIP6. The mei-yu/baiu/changma (WNP) rainfall anomalies are enhanced from 0.13 mm day− 1 (-0.73 mm day− 1) in the CMIP5 MME to 0.23 mm day− 1 (-0.87 mm day− 1) in the CMIP6 MME. Therefore, compared to the CMIP5 models, the simulation of interannual variability of EASM is improved in CMIP6. Among the CMIP6 models, it is worthy noting that the CESM2 model shows a high capability in reproducing the climatological wind and precipitation of EASM (Fig. 4b), and simulates reasonably the magnitudes of WNPACI (-1.91 s− 1) and rainfall anomalies over both the mei-yu/baiu/changma (0.35 mm day− 1) and WNP (-1.15 mm day− 1) regions (Fig. 7b).

4.1 Roles of simultaneous NIO and TNA SST anomalies

One issue that needs to be addressed is why the variability of EASM has been improved significantly from the CMIP5 to CMIP6 models? The reason may be rooted in the differences of SST anomalies in the tropics between the CMIP5 and CMIP6 models (e.g., Zhang et al. 1996; Wu et al. 2003; Xie et al. 2009; Rong et al. 2010; Wang et al. 2017; Yang and Huang2020; Chen et al. 2021). Hence, we first examine whether the models’ reproducibility of the observed anomalous WNPAC is related to the simultaneous SST anomalies in the CMIP5 and CMIP6 models.

Figure 8 displays the SST anomalies regressed upon the EASMI in the observation and the MMEs of CMIP5 and CMIP6. The 200-hPa velocity potential and divergent winds are also employed to examine the Walker circulation anomalies induced by anomalous SST (Fig. 9). In the observation, Pacific SST anomalies associated with a stronger EASM show pronounced warming anomalies confined in the tropical eastern Pacific and western Pacific, together with weak cooling anomalies in the tropical central Pacific (Fig. 8a). Meanwhile, there are significant warm SST anomalies in both the NIO (0°–20°N, 50°E–100°E) and TNA (0°–20°N, 80°W–20°W) regions (Fig. 8a). In response to the abovementioned SST anomalies, the rising part of the Walker circulation tends to shift westward, thus reducing convection over the western to central Pacific and increasing convection over the tropical eastern Indian Ocean (Fig. 9a). The suppressed convection may strengthen directly the anticyclonic vorticity anomaly over the WNP by emanation of descending Rossby waves (Fig. 6a; Gill 1980). The warming SST anomalies in both the NIO and TNA (Fig. 8a) may play important roles in maintaining the anomalous anticyclone over the WNP, too. In response to the NIO warming, troposphere temperature displays a Matsuno–Gill pattern and a Kelvin wave wedge penetrates the western Pacific along the equator, with a pronounced anticyclonic circulation on its northern flank (Yang et al. 2007; Xie et al. 2009). The TNA warming induces an east-west overturning vertical circulation with ascending motion over the tropical Atlantic and descending motion over the tropical central Pacific (Fig. 9a). This anomalous descending motion can reinforce the low-level anticyclonic anomaly to the west and therefore strengthen the EASM (Fig. 9a; Hong et al. 2014; Wang et al. 2017; Chen et al. 2018). In addition, the TNA warming can also enhance the WNPAC remotely through yielding Kelvin waves propagating eastward across Indian Ocean to the WNP (Rong et al. 2010; Ham et al. 2013; Chen et al. 2020). Takaya et al. (2021) suggests that the former route may contribute more appreciable influence on the EASM compared with the latter one.

Fig. 8
figure 8

Regression of JJA SST anomalies (°C) upon the EASMI in (a) observation, (b) CMIP5 MME, and (c) CMIP6 MME. Black dots indicate SST anomalies exceeding the 90% confidence level. The boxes shown in these figures are the NIO (0°–20°N, 50°E–100°E) and TNA (0°–20°N, 80°W–20°W) regions, respectively

Full size image
Fig. 9
figure 9

Regression of JJA 200-hPa velocity potential (contour; 105 m2s− 1) and divergent wind (vectors; ms− 1) upon the EASMI in (a) observation, (b) CMIP5 MME, and (c) CMIP6 MME. The contour interval is 5 × 105 m2 s− 1. Shading indicates anomalies exceeding the 90% confidence level

Full size image

In comparison, the central Pacific cooling associated with the EASM in the CMIP5 and CMIP6 MMEs becomes much stronger and extends to cover nearly the entire tropical Pacific (Fig. 8b, c), which is a common bias among coupled models (Song and Zhou 2014b). In addition, the NIO warming are very weak and the TNA SST anomalies are almost absent in the CMIP5 MME (Fig. 8b). In this case, the sinking branch of anomalous Walker circulation shrinks westward into the western Pacific (Fig. 9b). Thus, the suppressed convection anomalies over the WNP become weak, and the associated WNPAC becomes weak, too (Fig. 6b). However, such biases seem to be improved significantly in the CMIP6 models with the comparable NIO and TNA warming to the observation in the CMIP6 MME (Fig. 8c). And the sinking branch of the anomalous Walker circulation moves eastward into the central-western Pacific with a pattern quite consistent with the observation (Fig. 9a, c). Correspondingly, the western North Pacific suppressed convection and the associated anomalous WNPAC are also well simulated (Fig. 6c). In order to further examine roles of the NIO and TNA SST anomalies in the improvement of EASM simulation, scatterplots of the summertime NIO and TNA SST anomalies versus the WNPACI are shown in Fig. 10. The results indicate that the larger warm SST anomaly in both the NIO and the TNA tends to reproduce a more realistic anomalous WNPAC in the CMIP6 models, with the inter-model correlation coefficients of 0.52 and 0.46 (both exceeding the 95% confidence level), respectively (Fig. 10c, d). In contrast, the magnitudes of NIO and TNA SST anomalies in the CMIP5 models (Fig. 10a, b) are generally small compared to those in the CMIP6 models (Fig. 10c, d). Hence, the inter-model correlations of the anomalous WNPAC with the NIO and TNA SST anomalies are only 0.13 and 0.15, respectively (Fig. 10a, b). Previously studies have paid much attention to the importance of simulating Indian Ocean SST, which is vital to predict the WNPAC related to the EASM (e.g., Song and Zhou 2014a; Xin et al. 2020). Based on the above results, however, the accurate simulation of TNA SST seems to be equally important. Therefore, the model’s reproducibility of the observed anomalous WNPAC may be closely tied to the accurate simulation of simultaneous warm SST anomalies in the NIO and TNA.

Fig. 10
figure 10

Scatterplots of the JJA WNPACI vs. (a) NIO and (b) TNA SST anomalies regressed upon the EASMI in the 18 CMIP5 models. (c) and (d) same as in Figures (a) and (b) but for CMIP6 models. Black lines and “r” denote linear fit

Full size image

4.2 Role of preceding winter ENSO

The above analysis shows that the summertime SST anomalies in the NIO and TNA play important roles in reasonable simulation of interannual variability of EASM in the CMIP models. Previous studies have found that the NIO and TNA warming are closely tied to the preceding El Niño (Alexander et al. 2002 Chang et al. 2006; Xie et al. 2009). Hence, Fig. 11 shows the evolution of SST anomalies related to the EASM from the preceding winter to the following summer in the observation, CMIP5 and CMIP6. In the observation, the winter SST anomalies show a typical ENSO pattern, in which significant warm SST anomalies are observed in the tropical central and eastern Pacific (Fig. 11a). Such ENSO pattern in the Pacific begins to decay in spring (Fig. 11b), and weak cold SST anomalies occur in the equatorial central Pacific in summer (Fig. 11c). This evolution actually reflects decaying of an El Niño event. In addition, significant warm SST anomalies are observed in both the NIO and TNA, which can maintain from winter to the following summer (Fig. 11a–c).

Fig. 11
figure 11

Regression of SST anomalies (°C) evolution during the (a) DJF, (b) MAM and (c) JJA upon the EASMI in the observation. (d)–(f), (g)–(i) same as in Figures (a)–(c) but for CMIP5 MME and CMIP6 MME, respectively. Black dots indicate SST anomalies exceeding the 90% confidence level

Full size image

The CMIP5 MME fails to reproduce the observed ENSO-like SST anomalies in the preceding winter. Specifically, there is central Pacific SST cooling persisting from spring to the following summer (Fig. 11e and f). Correspondingly, there is faint warming SST in the NIO and TNA (Fig. 11d–f). Thus, the resultant atmospheric responses are weak in summer (Fig. 9b), and a weak WNPAC is simulated in the CMIP5 MME (Fig. 6b). However, in the MME of CMIP6 models, the SST anomalies in the preceding winter resemble the observed ENSO pattern quite well with a somewhat smaller magnitude than that in the observation (Fig. 11 g). In addition, the NIO and TNA warmings are also comparable to the observation, and the evolution of SST anomalies is generally simulated well, too (Fig. 11 g–i). Hence, the results imply that the reproducibility of ENSO SST pattern may be important to the formation of NIO and TNA SST anomalies in the CMIP models.

Equation (1) is applied to evaluate the simulation skill of ENSO SST pattern in the individual CMIP5 and CMIP6 models, and the ENSO SST pattern is defined as the SST anomalies in the tropical and subtropical Pacific (20°S–20°N, 120°E–80°W) regressed against the normalized Niño-3.4 index. The CMIP6 models generally perform well in the simulation of ENSO SST pattern than the CMIP5 models with the skill score 0.79 (0.66) for the CMIP6 (CMIP5) MME (Fig. 12). In addition, there exists a larger diversity of the skill score among the CMIP5 models (Fig. 12a) than the CMIP6 models (Fig. 12b). Previous studies have indicated that there is excessive westward extension of ENSO SST anomalies in the CMIP5 models (e.g., Jiang et al. 2021; Gong et al. 2015). The evolution of SST anomalies related to ENSO are also checked and this bias of excessive westward extension of ENSO SST anomalies in the CMIP5 models has been overcome greatly in the CMIP6 models (Figures not shown). Figure 13 further shows the scatterplots of the ENSO skill scores against the NIO and TNA SST anomalies. The inter-model correlation coefficients in the CMIP6 models are high and significant at the 95% confidence level (Fig. 13c, d). In contrast, the relationships between ENSO and the NIO/TNA SST anomalies are weaker in CMIP5 (Fig. 13a, b) comparing to CMIP6. The results suggest that a model with high ENSO simulation skill tends to produce realistic warm SST anomalies in the NIO and TNA in the following summer. Hence, the simulated SST links between ENSO and the following summer NIO/TNA may contribute to improvement in the EASM simulation on the interannual time scale.

Fig. 12
figure 12

Skill scores of DJF SST with respect to Niño-3.4 index in tropical eastern Pacific (20°S–20°N, 120°E–80°W) in (a) CMIP5 and (b) CMIP6 models, respectively. The dashed lines indicate the skill score of MME

Full size image
Fig. 13
figure 13

Scatterplots of the ENSO skill score vs. JJA (a) NIO and (b) TNA SST anomalies regressed upon the EASMI in the 18 CMIP5 models. (c) and (d) same as in Figures (a) and (b) but for CMIP6 models. Black lines and “r” denote linear fit

Full size image

Figure 14 examines the difference of SST evolution regressed upon the Niño-3.4 index between the CMIP6 and CMIP5 MMEs. In winter, the differences appear mainly in the tropical Pacific, indicating a better SST simulation of ENSO pattern in CMIP6 than CMIP5 (Fig. 14a). However, the signals become stronger in the NIO and TNA regions in the following seasons, especially in summer (Fig. 14b, c). The precipitation and 850-hPa winds over East Asia regressed on the Niño-3.4 index are also checked (Figures not shown), and the results show similar patterns to those regressed on the EASMI as shown in Fig. 6. Therefore, evolutions of ENSO and the associated SST anomalies in the NIO and TNA regions play an important role in the EASM simulation. A better simulated ENSO may provide a favorable background for the evolutions of NIO and TNA SST anomalies, which tends to contribute a better simulation of the anomalous WNPAC. Then a realistic anomalous EASM may eventually be simulated together with the precipitation and circulation variability over East Asia.

Fig. 14
figure 14

Regression of SST anomalies (°C) evolution during the (a) DJF, (b) MAM and (c) JJA upon the Niño-3.4 index in the difference between CMIP6 MME and CMIP5 MME. Black dots indicate SST anomalies exceeding the 90% confidence level. The boxes shown in these figures are the NIO (0°–20°N, 50°E–100°E) and TNA (0°–20°N, 80°W–20°W) regions, respectively

Full size image

5 Summary

The climatology and interannual variability of EASM have been evaluated based on 18 CMIP6 models and then compared to those in the CMIP5 models. It is found that the simulation capability of EASM has been improved in the latest CMIP6 models. Specifically, the CMIP6 models significantly improve the simulation of two key systems associated with the EASM: the WNPAC and mei-yu/baiu/changma rainband. For the EASM climatology, the performances of models are quantitatively measured by the Taylor diagram and skill score formula. The results indicate that the CMIP6 MME shows more skillful than the CMIP5 MME in both the climatological circulation and precipitation in summer over East Asia. The skill scores of U850, V850, Pr are significantly enhanced from the CMIP5 MME (0.62; 0.44; 0.49) to the CMIP6 MME (0.7; 0.56; 0.59) over East Asia. In addition, the biases of the WPSH and associated rainfall centers are reduced, and the inter-model spreads are smaller in the CMIP6 models compared to those in the CMIP5 models.

For the interannual variability of EASM, the precipitation and 850-hPa wind regressed upon the EASMI are examined. The results indicate that both the biases of the WNPAC and the East Asian dipole rainfall pattern are alleviated in the CMIP6 MME comparing to the CMIP5 MME. The WNPAC is strengthened in the CMIP6 MME (-1.22 s− 1) compared to that in the CMIP5 MME (-0.98 s− 1), which is more closer to the observation (-1.36 s− 1). Correspondingly, there are enhanced rainfall anomalies over the mei-yu/baiu/changma region (0.13 mm day− 1 in CMIP5 to 0.23 mm day− 1 in CMIP6) and the WNP (-0.73 mm day− 1 in CMIP5 to -0.87 mm day− 1 in CMIP6).

As expected, the improvement of the interannual variation of EASM in the CMIP6 models can be traced back to the improved tropical SST anomalies. The simultaneous central Pacific cooling is well simulated in models, with commonly stronger intensity and larger domain compared with observation. In contrast, the reasonable simulation of NIO and TNA SST anomalies are found to be the key factors. The NIO warming is much weak and the TNA SST anomalies are almost absent in CMIP5 MME, while those are realistically simulated in the CMIP6 MME. The suppressed convection over the WNP and the associated anomalous WNPAC are well reproduced if significant warm SST anomalies are seen in both the NIO and TNA in CMIP models. The summertime NIO and TNA warming are closely tied to the preceding ENSO. The CMIP6 MME performs high skill in the simulation of ENSO SST pattern compared with CMIP5. More realistic ENSO SST pattern may provide a favorable background for the evolutions of NIO and TNA SST anomalies in models, which tends to contribute a more realistic WNPAC associated with EASM. Hence, the reasonable simulation of EASM interannual variation is revealed to be strongly dependent on ENSO and the resultant summertime NIO and TNA SST anomalies. The result indicates that more attention should be given to understanding and improving simulation of the inter-basin connections in climate models.

In this study, we mainly focus on the reproducibility of JJA EASM in CMIP models. Moreover, the EASM experiences remarkable seasonal evolution, whose simulation remains a great challenge (Ding 2007; Feng et al., 2014). Many studies have demonstrated that the seasonal evolution of EASM is closely related to the climate mean status (Huang et al., 2003; Ding, 2007; Ren et al., 2013). The present results indicate that the CMIP6 models improve from the CMIP5 models in the climatology of EASM. Our preliminary analysis suggests that this improvement may be also rooted in the tropical SST. The reproducibility of the EASM seasonal evolution in models and the possible factors need to be explored in the near future.